[Beowulf] Nehalem and Shanghai code performance for our rzf example

[Nifty Tom Mitchell]

On Mon, Jan 19, 2009 at 09:03:23PM -0800, Karen Shaeffer wrote:
> On Mon, Jan 19, 2009 at 05:18:19PM -0500, Joe Landman wrote:
> > >
> > These days, perception == reality :(
> 
> Actually, Jeane Kirkpatrick coined the phrase, "Perception is more
> important than reality." But then that can only be held true in
> a contrived environment, because, in the end, "Reality isn't optional".
> 
> As for GaAs, it is not suitable for large scale integration or low
> power, both essential requisites to displace Si.
> 

One item to add to this discussion is the fan out of ECL logic which is
so much wider than TTL logic.  ECL can be built with both silicon and
with GaAs transistors.   In one discussion I think ECL/GaAs was compared to Silicon CMOS.
With the vast transistor counts of modern CMOS processors and high speed
on chip switching speed there is less need for high device to device 
fan out of ECL logic.

Silicon CMOS logic has a power advantage...

GaAs in some cases has a three fold speed advantage.

http://books.google.com/books?id=QHtalNXHKbsC&pg=PA450&lpg=PA450&dq=GaAs+versus+Silicon+switching+speed&source=web&ots=G0Ysy1kiAw&sig=QQBFInUFAOj5UKOjhqjn--GWQyM&hl=en&sa=X&oi=book_result&resnum=10&ct=result

Once the door to exotics is open,  diamond may be interesting.
	http://www.theinquirer.net/inquirer/news/396/1044396/diamond-transistor-clocks
	http://www.nanotech-now.com/news.cgi?story_id=27984

The thermal properties of diamond are compelling... 

S. Cray used GaAs and some interesting interconnect and cooling tricks
on the CrayIII project if I recall.   The transistor counts on
Si CMOS have effectively erased any potential advantages of ECL on practical
systems.    



-- 
	Regards,
	T o m   M i t c h e l l